import tensorflow as tf
from tensorflow.keras import layers

import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
import pandas as pd
import os

tf.__version__

'2.5.0'

path = './alfons_mucha_winter_1896.jpg'
original_img = tf.keras.preprocessing.image.load_img(path)
original_img

img_height, img_width, channels = tf.keras.preprocessing.image.img_to_array(original_img).shape
print(img_height, img_width, channels)

600 311 3

tf.expand_dims(original_img, 0)[0].shape

TensorShape([600, 311, 3])

def display_augmentation(image, data_augmentation=None, cmap=None):
    
    # change 3D-Tensor to 4D of shape=(1, 600, 311, 3)
    image = tf.expand_dims(image, 0)

    plt.figure(figsize=(15, 15))

    for i in range(9):
        
        # augment image
        augmented_image = data_augmentation(image)
        
        ax = plt.subplot(3, 3, i + 1)
    
        plt.imshow(augmented_image[0], cmap)

        plt.axis("off")

1. Random Flip

Randomly flip each image horizontally and/or vertically.

Horizontal + Vertical

from tensorflow.keras.layers.experimental.preprocessing import RandomFlip

image = original_img

data_augmentation = tf.keras.Sequential([
  RandomFlip("horizontal_and_vertical"),
])

display_augmentation(image, data_augmentation)

Vertical

image = original_img

data_augmentation = tf.keras.Sequential([
  RandomFlip("vertical"),
])

display_augmentation(image, data_augmentation)

Horizontal

image = original_img

data_augmentation = tf.keras.Sequential([
  RandomFlip("horizontal"),
])

display_augmentation(image, data_augmentation)

2. Random Rotation

Randomly rotate each image.

from tensorflow.keras.layers.experimental.preprocessing import RandomRotation

image = original_img

data_augmentation = tf.keras.Sequential([
  RandomRotation(0.3),
])

display_augmentation(image, data_augmentation)

3. Random Crop

Randomly crop the images to target height and width.

from tensorflow.keras.layers.experimental.preprocessing import RandomCrop

image = original_img

data_augmentation = tf.keras.Sequential([
  RandomCrop(int(img_height/2), int(img_width/2)),
])

display_augmentation(image, data_augmentation)

4. Random Contrast

Adjust the contrast of an image or images by a random factor.

from tensorflow.keras.layers.experimental.preprocessing import RandomContrast

image = original_img

data_augmentation = tf.keras.Sequential([
  RandomContrast(0.7),
])

display_augmentation(image, data_augmentation)

5. Random Height

Randomly vary the height of an image or images.

from tensorflow.keras.layers.experimental.preprocessing import RandomHeight, Rescaling

image = original_img

data_augmentation = tf.keras.Sequential([
    Rescaling(1./255),
    RandomHeight(factor=(-0.8, 0.8))
])

display_augmentation(image, data_augmentation)

6. Random Translation

Randomly translate each image during training.

from tensorflow.keras.layers.experimental.preprocessing import RandomTranslation

image = original_img

height_factor = (-0.5, 0.5)
width_factor = (-0.5, 0.5)

data_augmentation = tf.keras.Sequential([
  RandomTranslation(height_factor=height_factor,
                    width_factor=width_factor),
])

display_augmentation(image, data_augmentation)

7. Random Width

Randomly vary the width.

from tensorflow.keras.layers.experimental.preprocessing import RandomWidth

image = original_img

data_augmentation = tf.keras.Sequential([
    Rescaling(1./255),
    RandomWidth((-0.8, 0.8))
])

display_augmentation(image, data_augmentation)

8. Random Zoom

Randomly zoom each image.

from tensorflow.keras.layers.experimental.preprocessing import RandomZoom

image = original_img

height_factor = (-0.5, 0.5)
width_factor = (-0.5, 0.5)

data_augmentation = tf.keras.Sequential([
  RandomZoom(height_factor=height_factor,
             width_factor=width_factor)
])

display_augmentation(image, data_augmentation)

9. RGB to Grayscale

Converts one or more images from RGB to Grayscale

from tensorflow.image import rgb_to_grayscale

image = rgb_to_grayscale(original_img)

print(f'Number of channels = {image.shape[2]}')

plt.imshow(image, cmap='gray')

Number of channels = 1

<matplotlib.image.AxesImage at 0x7fee583ed1c0>

10. Random Saturation

Adjust the saturation of RGB images by a random factor.

plt.figure(figsize=(15, 15))

for i in range(9):

    # augment image
    augmented_image = tf.image.random_saturation(original_img, 0, 3)

    ax = plt.subplot(3, 3, i + 1)

    plt.imshow(augmented_image)

    plt.axis("off")

os.system('jupyter nbconvert --to html Tabaka_Marcin_shapes_of_winter.ipynb')

0